Inverse Reinforcement Learning for Personalized and Safe Trajectory Planning in Autonomous Driving

To enable human-like autonomous driving, we propose a framework that integrates personalization with safety. It first establishes safe trajectory boundaries using a multi-factor coupled field model. Then, it quantifies driving style by learning reward weights from driver decisions via Maximum Entropy Inverse Reinforcement Learning (MaxEnt IRL). Finally, these learned parameters guide polynomial trajectory planning within the safety constraints. Experiments validate the method's effectiveness in generating safe, stable, and personalized path planning.